Cannon-Brookes on Imagination Asymmetry and Atlassian's Graph

Key Takeaways

• Didier Elzinga, the founder of Culture Amp, defined the shift from information asymmetry to imagination asymmetry, which Mike Cannon-Brookes calls the defining test of defensibility.
• Possessing proprietary business facts no longer protects a company because raw knowledge has become cheap, accessible software commodity.
• Atlassian designs its Teamwork Graph to function like Google's index of the web, predicting what an AI agent requires before that agent queries the database.
• Enterprise context requires dual tracking: hyper-fresh operational data from yesterday alongside compressed historical facts, such as Cannon-Brookes coding Java in 2003.

The Death of Information Asymmetry

For thirty years, enterprise software companies built moats out of proprietary data caches. If you knew supplier pricing, buyer history, or technical specifications that rivals lacked, you won. Cannon-Brookes argues that era ended with modern language models.

Borrowing a concept from Culture Amp founder Didier Elzinga, Cannon-Brookes points to a reset in how companies compete: “We're moving from a world of information asymmetry to imagination asymmetry, which is such a great way of thinking about it, right?”▶ 35:19

When any competitor can query a foundation model for market strategy, code architectures, or legal summaries, possessing static facts provides zero protection. Cannon-Brookes put it plainly: “It's no longer I know something that someone else doesn't or I have a particular business knowledge. those knowledges are becoming less about the asymmetric value than can I think about how to apply AI and these technologies to problems in unique ways.”▶ 35:27

Building Google's Index for Internal Work

If raw models flatten external knowledge, defensibility moves inside company walls. The battleground is how software models team activity, decisions, and workflows in real time.

Atlassian built its Teamwork Graph to map how people, tickets, code commits, and documents interact. Autonomous agents fail inside enterprises when they lack live operational context. If an agent answers a question by searching raw documents from scratch, it hallucinates or acts on stale data.

Cannon-Brookes compares Atlassian's approach to consumer search: “The teamwork graph's job is to infer ahead of your agent before your agent has asked a question to inferred the best possible answer to that. Think about it like Google's index of the web.”▶ 40:18 The graph pre-calculates relationships so software acts on structured truth rather than guessing context on the fly.

What to Remember and What to Forget

A common mistake among founders building internal knowledge graphs is storing every event with equal priority. Dumping twenty years of git logs, chat logs, and closed tickets into a retrieval pipeline drowns an AI agent in noise.

Cannon-Brookes uses his own engineering history at Atlassian to show the difference between raw noise and useful context: “Knowing that I wrote Java code in 2003 and what code I wrote is totally useless to Atlassian in 2026. Knowing the fact that I used to be a programmer, a bad one, 20 plus years ago, is kind of useful.”▶ 41:51

Winning systems separate high-level identity from volatile operational reality. “An organizational graph has to be very fresh and current with facts and very historical in its learnings and memories to help the organization as best as possible,”▶ 42:39 Cannon-Brookes explained. Stale facts must expire fast. Enduring roles, skills, and cultural norms stay permanent.

What to Do With This

Audit your internal context pipeline before deploying your next internal agent. Write down the top five data sources feeding the system, then set a strict expiration date on tactical work items older than 90 days. Keep the permanent metadata about who holds expertise, but delete the line-by-line task logs from three years ago so your models stop retrieving obsolete context.